Air pressure reducing valve fault diagnosis method based on feature fusion

Abstract

The invention provides an air pressure reducing valve fault diagnosis method based on feature fusion, and relates to the field of aviation element fault diagnosis, and the method comprises the following steps: S1, screening out performance degradation key parameters; s2, key parameter characteristics of the starting process and the steady state process of the air pressure reducing valve are extracted; s3, selecting characteristic parameters in a healthy state, and constructing a characteristic matrix based on the characteristic parameters; s4, performing standardization processing on each column of the feature matrix to form a whitening parameter matrix; s5, performing derivation on the whitening parameter matrix according to column vectors to form a first-order derivative matrix; S6, obtaining a fused feature parameter matrix; and S7, calculating a three-time standard deviation of the target feature vector as a fault boundary threshold. The fault diagnosis method is provided by fusing multiple features, the fault boundary threshold is dynamically updated based on a small number of healthy samples through a self-adaptive fault threshold confirmation mechanism, the method can effectively recognize the early fault of the air pressure reducing valve, and the misinformation and missing report risks are remarkably reduced.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis of aerospace components, and specifically to a fault diagnosis method for air pressure reducing valves based on feature fusion. Background Technology

[0002] An air pressure reducing valve is a device that automatically reduces the working pressure of a pipeline. Through closed-loop regulation, it reduces the higher gas pressure upstream of the valve to the required level downstream, and relies on the energy of the medium itself to automatically maintain a stable outlet pressure. From a fluid mechanics perspective, a pressure reducing valve is a throttling element with variable local resistance. That is, by changing the throttling area, the flow velocity and kinetic energy of the fluid are altered, resulting in different pressure losses, thereby achieving pressure reduction. Then, through the regulation of the control and adjustment system, the fluctuations in downstream pressure are balanced with the spring force, maintaining a constant pressure within a certain error range.

[0003] Air pressure reducing valves are widely used, especially in critical systems of aircraft, ships, and various industrial equipment. Early warning of pressure reducing valve failures, early detection of faults, and replacement of pressure reducing valves are key means to improve system reliability and ensure the continuous operation of critical systems.

[0004] Factors such as valve core jamming, spring assembly aging, and seal ring wear and failure can all cause abnormalities in air pressure reducing valves. If such faults are not detected in time, they can easily induce other types of faults and even lead to unplanned shutdowns of the entire system, seriously affecting equipment use and causing unnecessary economic losses.

[0005] Currently, traditional fault monitoring methods rely on simple upper and lower limit readings of key parameters based on usage status and single-use data. This method has low monitoring costs and can quickly and intuitively detect abnormalities in the operation of the pressure reducing valve. However, this monitoring method is difficult to effectively capture early, subtle fault characteristics, thus failing to detect early faults in the pressure reducing valve. This results in poor system adaptability and makes predictive maintenance impossible. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to provide a fault diagnosis method for air pressure reducing valves based on feature fusion.

[0007] Specifically, the present invention provides a fault diagnosis method for air pressure reducing valves based on feature fusion, which includes the following steps: S1: Based on the working principle of the air pressure reducing valve and the typical fault phenomena and fault modes it exhibits during use, the key parameters related to its performance degradation are selected. The key parameters include inlet gas pressure, outlet gas pressure and temperature. S2: For the key parameters selected in S1, extract the key parameter features of the two stages of the air pressure reducing valve startup process and steady-state process respectively; S3: Based on the key parameter features extracted in S2, select feature parameters of any device in its health state, and construct a feature matrix A based on these feature parameters. The row vectors of the feature matrix represent the key parameter features, and the column vectors represent the device sequence. ; Where i represents the current device, j represents the total number of flights, and p represents the characteristic parameters. Corresponding to a set of health status feature vectors; S4: Standardize each column of the feature matrix constructed in S3 to form the whitening parameter matrix B. ; S5: Take the derivative of the whitening parameter matrix formed in S4 using column vectors to form the first derivative matrix D: ; Then, the covariance matrix of the first derivative matrix is ​​calculated and the covariance matrix is ​​decomposed to obtain the eigenvalues ​​of the derivative matrix and their corresponding eigenvectors. The eigenvalues ​​are sorted from smallest to largest, and the direction of slowest change is selected as the target eigenvector. S6: Combine the whitening parameter matrix formed in S4 and the target feature vector obtained in S5 to obtain the fused feature parameter matrix; S7: For the target feature vector obtained in S6, use the first t usage data of the air pressure reducing valve as the basic calibration data, where t≥50; calculate the standard deviation of the target feature vector based on the basic calibration data, and take 3 times the standard deviation as the fault boundary threshold. If the fused feature parameters in S6 exceed the fault boundary threshold, it is determined that there is a fault.

[0008] Furthermore, in S2, key parameter features include mean, variance, peak factor, and waveform factor.

[0009] Furthermore, in S4, the standardization process includes the Z-score method, which is used to eliminate the influence of dimensions on the dataset and make the dataset conform to a standard normal distribution.

[0010] Further: The standardization process using Z-scores involves calculating the mean and variance of the column vectors of the feature matrix constructed in S3, column by column, and then calculating the standard scores of the column vectors. ; in, Standard scores, The original data, Mean for each column vector, The variance for each column vector; Each data point in the feature matrix is ​​replaced with a calculated standard score to form a whitening parameter matrix.

[0011] Furthermore: In S5, the steps for taking the first derivative matrix by column vectors of the whitening parameter matrix are as follows: Solve for the first derivative column by column of the whitening parameter matrix: ; in, The first derivative of the standard score. To calculate the gradient coefficients for differentiation, in some preferred embodiments, Take 1; Then, each data point in the whitening parameter matrix is ​​replaced with the calculated first derivative.

[0012] Further: In S6, the method for obtaining the time series vector includes: multiplying the whitening parameter matrix formed in S4 with the target feature vector obtained in S5 to obtain the time series vector, and selecting the first dimension as the target feature vector.

[0013] Furthermore: In S7, twice the standard deviation is taken as the warning boundary threshold. When the fused feature parameters in S6 exceed the warning boundary threshold, a fault warning is issued for the air pressure reducing valve.

[0014] Preferably, the air pressure reducing valve is used in the aviation field.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: To address the issues that the fault status of air pressure reducing valves in aviation can only be monitored during engine aerodynamic processes, and that fault diagnosis information based on a single parameter is incomplete, this invention proposes a fault diagnosis method that integrates multiple features. By standardizing the feature matrix to form a whitening parameter matrix, the influence of dimensions on the dataset is eliminated, ensuring the dataset conforms to a standard normal distribution for easier statistical analysis. By differentiating the whitening parameter matrix and obtaining the target feature vector, the instantaneous change trend of feature parameters with usage frequency can be captured, effectively amplifying the subtle and continuous changes exhibited by the pressure reducing valve in early-stage faults, making the early warning model more sensitive to performance degradation. Furthermore, by combining the whitening parameter matrix and the target feature vector, fused feature parameters are obtained. These fused feature parameters are then compared with an inductively derived fault boundary threshold to diagnose whether the air pressure reducing valve is faulty. This method effectively identifies early-stage faults in air pressure reducing valves, significantly reducing the risk of false alarms and missed alarms.

[0016] This invention addresses the problem that fixed threshold methods cannot adapt to dynamic operating conditions. It proposes an adaptive fault threshold confirmation mechanism that dynamically updates the fault boundary threshold based on a small number of healthy samples. To address the inconsistency in equipment installation and usage conditions, a dynamic threshold based on normal samples is proposed, which can more accurately predict the degradation trend of pressure reducing valves and effectively reduce the false alarm rate of fault warnings. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall process of the air pressure reducing valve fault diagnosis method based on feature fusion according to the present invention. Figure 2 This is a schematic diagram illustrating the working principle of the air pressure reducing valve of the present invention. Figure 3 This is a diagram illustrating the effect of the feature fusion-based fault diagnosis and early warning system for pressure reducing valves according to the present invention. Detailed Implementation

[0018] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0019] Combination Figure 1 and Figure 2 As shown. An air pressure reducing valve is a specialized device that automatically reduces the working pressure of a pipeline. It can reduce the high gas pressure in the pipeline before the valve to the required level in the pipeline after the valve. After air enters the air pressure reducing valve from high pressure, the high pressure air is reduced to the required pressure range, thereby regulating the air flow rate and pressure to maintain a suitable air-fuel mixture ratio in the combustion chamber and ensure successful ignition in the combustion chamber.

[0020] This invention proposes a fault diagnosis method for air pressure reducing valves based on feature fusion, which includes the following steps: S1: Identify key parameters related to the performance degradation of the air pressure reducing valve: Based on the working principle of the air pressure reducing valve and the typical fault phenomena and fault modes exhibited during its use, screen out key parameters related to its performance degradation. Key parameters include inlet gas pressure, outlet gas pressure, and temperature, etc. This is the basis for data acquisition and subsequent feature extraction, ensuring that subsequent analysis targets the core variables that can reflect the health status of the valve.

[0021] S2: Extracting Key Parameter Features of the Air Pressure Regulator: For the key parameters selected in S1, key parameter features are extracted for both the startup and steady-state stages of the air pressure regulator. These key parameter features include p statistical features such as mean, variance, peak factor, and waveform factor. This step comprehensively quantifies and describes the behavioral patterns of parameters as they change with usage frequency, providing rich raw information for feature fusion.

[0022] Taking the air pressure relief valve of the aircraft auxiliary power system as an example, let the feature set of the air pressure relief valve of the i-th aircraft for the j-th flight be . Each feature set contains p features. , This represents the t-th feature of the key parameter of the air pressure relief valve of the i-th aircraft in the j-th sortie.

[0023] S3: Construct the feature parameter matrix of key parameters of the air pressure reducing valve under healthy conditions: Based on the key parameter features extracted in S2, select the feature parameters of any device i under healthy conditions. Construct a feature matrix A based on p feature parameters, where the row vectors of the feature matrix represent key parameter features and the column vectors represent device sequences: ; Where i represents the current device, j represents the total number of flights, and p represents the characteristic parameters. This corresponds to a set of health status feature vectors; each feature is counted from the start of use, and the sequence of usage counts forms a set of column vectors, which serve as a set of feature vectors. This matrix is ​​established to define the feature space when the device is healthy, and it forms the basis for subsequent feature fusion operations such as standardization and differentiation.

[0024] S4: Standardize the constructed feature parameter matrix to form a whitening parameter matrix: Since the numerical range of each column in the constructed feature parameter matrix is ​​not uniform, it belongs to a matrix with different characteristics for each column. To facilitate statistical analysis, the feature matrix constructed in S3 is standardized to obtain a whitening matrix. The standardization calculation adopts the Z-score method. The purpose of standardization is to eliminate the influence of the dimensions of the dataset, so that the dataset conforms to the standard normal distribution, which facilitates subsequent differentiation operations and feature fusion. The standardization formula is as follows: ; in, It is the raw data. It is the mean calculated for each column vector. This is the variance calculated for each column vector. The final whitening matrix B is shown below: .

[0025] S5: Differentiate the resulting whitening matrix by column vectors, calculate the covariance matrix of the first derivative matrix, solve for the eigenvectors and eigenvalues ​​of the covariance matrix, and sort them in ascending order; Differentiate the whitening parameter matrix formed in S4 by column vectors to form the first derivative matrix D: .

[0026] Then, the covariance matrix of the first derivative matrix is ​​calculated and the covariance matrix is ​​decomposed to obtain the eigenvalues ​​of the derivative matrix and their corresponding eigenvectors. The eigenvalues ​​are sorted from smallest to largest, and the direction of slowest change is selected as the target eigenvector. S6: Multiply the whitening parameter matrix with the sorted feature matrix to obtain the fused feature parameter matrix: Combine the whitening parameter matrix formed in S4 and the target feature vector obtained in S5 to obtain the fused feature parameter matrix; specifically, multiply the whitened matrix by the feature vector to obtain a new set of time series vectors, and select the first dimension as the target feature vector.

[0027] S7: Calculate the fault warning boundary threshold based on the newly generated feature vector: Due to significant differences in equipment installation and usage conditions, to more accurately predict the degradation trend of the pressure reducing valve, the data from the previous t uses of the pressure reducing valve are used as the basic calibration data, where t ≥ 50. Thresholds are determined empirically, with 2 times the standard deviation used as the warning boundary threshold and 3 times the standard deviation used as the fault boundary threshold, to provide fault warnings for the pressure reducing valve. If the fused feature parameters from S6 exceed the fault boundary threshold, a fault is determined to exist. This method addresses the inconsistency in equipment installation and usage conditions by proposing a dynamic threshold based on normal sample updates, which can more accurately predict the degradation trend of the pressure reducing valve.

[0028] The above steps are described in more detail below through examples: Example 1 Three key statistical features were selected (p=3), and data were collected from five usage flights (j=5) of the equipment in a healthy state.

[0029] Step 1 and Step 2: Determine key parameters and extract features: Extract the following 3 features (p=3): Mean; Variance; Peak factor.

[0030] Step 3: Construct the feature parameter matrix ( ).

[0031] A 5×3 feature matrix is ​​obtained. The row vector represents the equipment sequence, i.e., 5 flights, and the column vector represents the key feature parameters, i.e., 3 features.

[0032] .

[0033] The data for these 5 sorties are as follows: Flight 1: ; Flight 2: ; Flight 3: ); Flight 4: ; Flight 5: .

[0034] For matrix Z-score standardization is performed column-wise (i.e., for each feature) to eliminate dimensional differences and ensure that the feature data conforms to a standard normal distribution. The mean of each column is then calculated. and standard deviation Then, through the formula Obtain the whitening parameter matrix '.

[0035] After standardization, the matrix The following is an example: ; Select the minimum value to construct a one-dimensional feature vector. .

[0036] Select the first derivative matrix according to the row vector. The minimum value in each row represents the direction of the most severe performance degradation among all features in that particular flight change. This generates a new one-dimensional feature vector. : ; The first element, -2.584, is the minimum value in the first row, and so on.

[0037] Finally, using the newly generated one-dimensional feature vector To calculate the dynamic threshold: First, calculate the statistical value of the vector: mean ( ): -1.77.

[0038] Standard deviation ( ): 0.814.

[0039] Use 2 standard deviations and 3 standard deviations as the lower limit thresholds for warnings and faults, because The focus is on drastic negative changes: Early warning boundary threshold ( ): .

[0040] Fault boundary threshold (3) ): .

[0041] Conclusion: The dynamic threshold based on healthy samples is set as follows: (Early warning) and (Fault), that is, when a new usage flight's fused characteristic parameters are subsequently detected. Below When this happens, the system will issue an alarm signal; when new usage flight fusion characteristic parameters are subsequently detected... If the value is below -4.21, the air pressure reducing valve is considered to be faulty.

[0042] like Figure 3 The figure shown is a diagram of the fault diagnosis and early warning effect of pressure reducing valve based on feature fusion. The horizontal axis is the number of times it is used, the vertical axis is the standard deviation of the fused features, the green area is the healthy range, the pink area is the degradation range, and the part beyond the pink area is the fault range.

[0043] from Figure 3 As can be seen from the data, the 179th fault occurred at the actual point of failure. In fact, the pressure reducing valve had exceeded its limit at the 177th fault. After the replacement of the parts, the equipment entered the alarm zone and returned to the normal range after a break-in period.

[0044] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A fault diagnosis method for air pressure reducing valves based on feature fusion, characterized in that, It includes the following steps: S1: Based on the working principle of the air pressure reducing valve and the typical fault phenomena and fault modes it exhibits during use, the key parameters related to its performance degradation are selected. The key parameters include inlet gas pressure, outlet gas pressure and temperature. S2: For the key parameters selected in S1, extract the key parameter features of the two stages of the air pressure reducing valve startup process and steady-state process respectively; S3: Based on the key parameter features extracted in S2, select feature parameters of any device in its health state, and construct a feature matrix A based on these feature parameters. The row vectors of the feature matrix represent the key parameter features, and the column vectors represent the device sequence. ； Where i represents the current device, j represents the total number of flights, and p represents the characteristic parameters. Corresponding to a set of health status feature vectors; S4: Standardize each column of the feature matrix constructed in S3 to form the whitening parameter matrix B. ; S5: Take the derivative of the whitening parameter matrix formed in S4 using column vectors to form the first derivative matrix D: ； Then, the covariance matrix of the first derivative matrix is ​​calculated and the covariance matrix is ​​decomposed to obtain the eigenvalues ​​of the derivative matrix and their corresponding eigenvectors. The eigenvalues ​​are sorted from smallest to largest, and the direction of slowest change is selected as the target eigenvector. S6: Combine the whitening parameter matrix formed in S4 and the target feature vector obtained in S5 to obtain the fused feature parameter matrix; S7: Obtain the usage data of the air pressure reducing valve for the first t times as the basic calibration data, where t≥50; calculate the standard deviation of the target feature vector based on the basic calibration data, and take 3 times the standard deviation as the fault boundary threshold. If the feature parameters fused in S6 exceed the fault boundary threshold, it is determined that there is a fault.

2. The air pressure reducing valve fault diagnosis method based on feature fusion as described in claim 1, characterized in that: In S2, key parameter features include mean, variance, peak factor, and waveform factor.

3. The air pressure reducing valve fault diagnosis method based on feature fusion as described in claim 1, characterized in that: In S4, the standardization process includes the Z-score method, which is used to eliminate the influence of the dimensions of the dataset and make the dataset conform to a standard normal distribution.

4. The air pressure reducing valve fault diagnosis method based on feature fusion as described in claim 3, characterized in that: The steps for standardization using Z-scores are as follows: For the feature matrix constructed in S3, calculate the mean and variance of the column vectors column by column, and then calculate the standard scores of the column vectors: ； in, Standard scores, The original data, Mean for each column vector, The variance for each column vector; Each data point in the feature matrix is ​​replaced with a calculated standard score to form a whitening parameter matrix.

5. The air pressure reducing valve fault diagnosis method based on feature fusion as described in claim 1, characterized in that: In S5, the steps for taking the first derivative matrix by column vectors of the whitening parameter matrix are as follows: Solve for the first derivative column by column of the whitening parameter matrix: ； in, The first derivative of the standard score. The gradient coefficients for differentiation; Then, each data point in the whitening parameter matrix is ​​replaced with the calculated first derivative.

6. The air pressure reducing valve fault diagnosis method based on feature fusion as described in claim 1, characterized in that: In S6, the method for obtaining the time series vector includes: multiplying the whitening parameter matrix formed in S4 with the target feature vector obtained in S5 to obtain the time series vector, and selecting the first dimension as the target feature vector.

7. The air pressure reducing valve fault diagnosis method based on feature fusion as described in claim 1, characterized in that: In S7, twice the standard deviation is taken as the warning boundary threshold. When the fused feature parameters in S6 exceed the warning boundary threshold, a fault warning is issued for the air pressure reducing valve.

8. The air pressure reducing valve fault diagnosis method based on feature fusion as described in any one of claims 1-7, characterized in that: Air pressure reducing valves are used in the aviation industry.

Images